Using Artificial Intelligence to Automate the Analysis of Psoriasis Severity: A Pilot Study.

INTRODUCTION

The Psoriasis Area and Severity Index (PASI) score is widely used to assess psoriasis severity; however, manual PASI scoring is susceptible to environmental variability and subjective interpretation. This study leverages artificial intelligence to improve the consistency and objectivity of psoriasis severity classification based on features extracted from 2D clinical images.

METHODS

This study employed the YOLOv8 deep learning model to classify psoriatic lesions according to the severity of erythema, thickness, and scaling - key subcomponents of the PASI scoring system. Severity was assessed as follows: (0), mild (1), moderate (2), severe (3), or very severe (4). Model training and analysis were conducted in a cloud-based environment (Google Colab) using three different datasets. Stratified k-fold cross-validation was employed to ensure robustness by preserving the distribution of PASI scores across folds. Model performance was assessed using a confusion matrix and accuracy metrics.

RESULTS

In experiments, the YOLOv8 model proved highly effective in classifying psoriasis images based on PASI scores. Stratified k-fold cross-validation was shown to enhance model reliability across diverse datasets.

CONCLUSIONS

This study represents a significant advancement in the application of AI to the automated classification of lesion severity based on erythema, thickness, and scaling - key subcomponents of PASI.